Presented by the Huxiu Tech Team
Authors | Chen Yifan, Liu Xuanqi
Edited by Miao Zhengqing
Header image | AI-generated
This is the 21st article in the虎嗅WAIC series "Tracking New Token Business Paradigms"
In tech investing, age is never a sufficient indicator of ability. However, during phases of rapid technological paradigm shifts, young investors deserve separate consideration.
The reason is not complicated. AI, robotics, commercial spaceflight, and quantum computing are advancing simultaneously; the pace at which new technologies move from academic papers to products and from labs to industry has clearly accelerated. Traditional investment logic no longer applies. To participate in these projects, investors can no longer rely solely on financial models and experience from mature industries—they must also understand technical language, accept unconsolidated pathways, and make decisions before consensus emerges.
In the mobile internet era, a company could quickly prove itself through user growth, retention, revenue, and network effects; today’s AI and hard tech projects, however, may span algorithms, chips, sensors, supply chains, industrial applications, and regulatory systems. Investors must not only understand a technological leap but also assess whether it can become a stable product, establish a data and training feedback loop, and accompany the company through the long journey of engineering and commercialization.
At this year's WAIC, the "Alpha Youth Investment Leaders" list was unveiled, highlighting 10 next-generation investors. Yet beyond the list itself, what’s more significant is that when viewed together, these investment directions and judgments resemble a developing map of technology investment trends.
Details of the WAIC FUTURE TECH Young Investor Ranking
This chart reveals a common shift: capital is moving away from chasing technological buzzwords and toward identifying the real pathways through which technology enters industries. The next opportunity may not belong to the loudest concepts, but rather to areas where model红利 have yet to be unlocked, physical-world capabilities are still lacking, and infrastructure requires long-term capital to enter early.
Trend One: AI is leaving the screen and entering the physical world
Over the past few years, the primary entry points for market understanding of AI have been large models, chatbots, and generative applications. Today, investment focus is shifting toward edge devices, robots, automobiles, industrial systems, and space infrastructure. AI is no longer just about generating text or images—it is now about perceiving environments, understanding human intent, and taking actions in the real world.
This is also why embodied intelligence, humanoid robots, AI hardware, and edge-side intelligence are receiving concentrated attention. Zhou Xin, Executive Director of Jinqiu Fund, has been following robotics since 2018.
She reminded the roundtable that the bubble in embodied intelligence does not stem from a lack of technological progress, but from the market directly applying the rapid iteration pace of large language models to robots deeply integrated with the physical world. Improved model capabilities do not equate to simultaneous successful real-world deployment. A stunning demo can spark imagination, but it cannot answer whether a robot can reliably perform a task a hundred times in a row, whether its error rate is controllable, whether deployment and operational costs can be reduced, or whether the final ROI is viable.
This evaluation shifts the competition in embodied intelligence from "demo performance" back to "delivery capability." The real world doesn't offer retryable prompts. Factories, warehouses, homes, and public spaces demand systems that can handle noise, wear, unexpected events, and unpredictable human behavior. The model is just one layer—touch, sensing, actuators, control systems, data collection, and scenario feedback collectively determine whether a product works.
The热度 of capital can be quantified.
According to the Chinese Academy of Information and Communications Technology's "Development Report on Embodied Intelligence (2025)," as of December 2025, China witnessed 744 investment events in the field of embodied intelligence and robotics, with a total financing amount of RMB 73.543 billion. IDC predicts that global humanoid robot shipments will exceed 50,000 units in 2026, representing a year-over-year growth of 178%.
However, cold real-world adoption remains: in the first three quarters of 2025, 73.6% of Unitree Technology’s humanoid robot revenue came from the scientific research and education sectors. The equipment has been widely sold to universities and research institutions, but there is still a long way to go before it generates tangible value in factory settings. This precisely validates Zhou Xin’s warning: progress in models cannot be directly equated with progress in real-world deployment.
Edge-side intelligence has thus become another critical thread. Zhou Xin believes that if AI represents a cognitive revolution on par with the Industrial Revolution, intelligence cannot remain confined to the cloud indefinitely. The future cognitive network will be jointly composed of cloud and edge-side components: the foundation consists of chips and hardware platforms, the middle layer comprises execution environments centered around Agents, and the top layer features applications capable of understanding individual needs. Agents may gradually become the primary interaction interface for hardware, understanding users’ preferences, history, and status from above, while invoking devices such as smartphones, glasses, cars, robots, and home appliances from below.
This will transform the business model of AI hardware. Revenue will no longer come solely from one-time sales, but potentially from ongoing services that create lasting value. The true barrier to entry is no longer just “AI embedded in the device,” but whether the device can lower usage barriers and establish stable data and service relationships with users.
Several investors on the list have already aligned their strategies along this path: Bai Zeren, Vice President of Hou Xue Capital, focuses on autonomous last-mile delivery, haptic sensing, and autonomous trucking; Guo Jing, Investment Manager at Yaotu Capital, specializes in vertical-specific robots, drone swarms, and solar panel installation robots. Autonomous last-mile delivery is one of the fastest-moving sectors in “AI entering the physical world”: industry-wide funding approached RMB 10 billion in 2025. Bai Zeren’s firm, Linear Capital, invested in White Rhino, which completed three funding rounds that year, raising over $100 million total. The company’s active fleet grew from approximately 100 vehicles at the end of 2023 to over 2,000 by December 2025, operating routinely in more than 170 cities worldwide. These businesses lack flashy demos but are constantly tested on efficiency, cost, and reliability. After AI enters the physical world, technological imagination remains important—but scalable delivery will become the true dividing line.
Trend Two: The benefits of large models are not over; the focus of competition has shifted to the "intelligence flywheel."
While the market continuously seeks the next-generation architectures and new concepts, an easily overlooked fact is that the industrial benefits of this generation of large models may still be far from fully realized.
Hu Qi, Executive Director at Qiming Venture Partners, noted that one of today’s biggest false consensuses is the excessive focus on the next breakthrough technology, while insufficient attention is paid to how existing large models can truly drive productivity transformation. After TCP/IP became the standard protocol, the internet’s红利 took years to fully unfold; Bayesian theory and CNNs also continued to influence new technological systems many years after their initial proposal. The journey from the emergence of foundational technologies to industry-wide transformation is rarely a linear process spanning just one or two years. Large models have only entered the public eye in recent years, and business systems, organizational processes, and industrial infrastructure have yet to fully catch up with their capabilities.
This means investors don’t have to choose between “continuing to invest in large models” and “seeking the next generation of technology.” The more important question is which companies can integrate existing model capabilities into real-world workflows to create value that won’t be easily erased by the next model upgrade.
Liu Yunjie, Investment Director at Jingya Capital, refers to this capability as the "Intelligent Flywheel." Users generate data by using the product; this data feeds into training or optimization processes, enhancing model performance, which in turn improves the user experience. A better experience attracts more users, use cases, and feedback. The data loop, training loop, and user feedback loop mutually reinforce one another, enabling the company to achieve continuous self-enhancement.
When applying this framework to specific companies, Zhipu and Keling serve as two clear examples. During the panel, the host mentioned that Hu Qi had heavily invested in Zhipu prior to the market surge. Zhipu’s revenue grew from RMB 57.4 million in 2022 to RMB 312.4 million in 2024, achieving a three-year CAGR of approximately 130%. In the first half of 2025, revenue reached RMB 190.9 million, a 325% year-over-year increase, and in January 2026, it listed on the Hong Kong Stock Exchange, becoming the world’s first publicly traded company focused on large models. Meanwhile, Keling AI from Kuaishou demonstrated the rapid commercialization potential when model capabilities are integrated into creative workflows: it opened for beta testing in June 2024 and launched its paid subscription system within about 45 days. In full-year 2025, Keling generated approximately RMB 1.04 billion in revenue, with a single-month income exceeding $20 million in December. By the end of 2025, it had over 60 million global users, generated more than 600 million videos, and served over 30,000 enterprise customers and developers. User creation, data feedback, model iteration, and experience enhancement—each rotation of this flywheel directly translates into revenue.
From this framework, the key metrics for AI companies also need to change. ARR, GMV, and SaaS efficiency remain relevant, but they are insufficient to determine whether a company has long-term moats. A feature may rapidly lose value due to upgrades in the foundational model; however, a system that is truly integrated into business processes, capable of accumulating proprietary data, and continuously learning may become stronger as models advance.
Liu Yunjie further divides the intelligent value chain into three layers: generation of intelligence, distribution of intelligence, and embodiment of intelligence in the physical world. World models and drug discovery belong to the generation of intelligence; agents and enterprise-grade agents handle the distribution of intelligence; embodied intelligence and autonomous driving transform intelligence into physical actions. This framework breaks down the "AI赛道" into a complete value chain. Investors are no longer just seeking popular applications, but companies that can occupy platform positions and establish feedback loops at each layer.
Therefore, the focus of large model investment is shifting. The previous question was “Which model is stronger?”; the next question is “Who can turn model capabilities into a system that gets stronger with use and deeper into business operations?”
Trend Three: Data bottlenecks are driving the emergence of new underlying technologies, with scientific foundation models becoming a non-consensus direction.
As large models continue to expand, they will eventually confront a more fundamental issue: high-quality human data is limited, and collecting and annotating it is costly. When internet corpora and human feedback gradually approach their limits, where can models obtain new experiences?
Hu Qi views reinforcement learning, self-play, and continuous iteration as potential solutions. He cites AlphaGo Zero as an example: the model, knowing only the rules of Go, improves its capabilities through self-play-generated experience without relying on human game records. Whether similar mechanisms can extend from well-defined games to more complex scientific research, engineering systems, and real-world tasks will determine whether AI can break free from its dependence on existing data.
It also provides a set of criteria to evaluate the underlying technology: it must first drive widespread improvement across a sufficiently large industry, rather than benefiting only a narrow part of a single company; second, it must expand the boundaries of AI’s capabilities—for example, by addressing challenges in data, reasoning, or continuous learning; and finally, it must stand the test of time, not merely hold up during a single wave of technological hype.
Following this logic, scientific foundation models have emerged as a noteworthy non-consensus direction. They are not entirely the same as the commonly understood AI for Science, which is often interpreted as automated laboratories, protein structure prediction, material discovery, or isolated scientific tools; scientific foundation models aim to build more fundamental and generalizable model capabilities across fields such as life sciences, materials, and simulation, serving multiple tasks within a discipline.
The challenges in this direction are also clear. Scientific data is characterized by high specialization, inconsistent standards, and costly acquisition, and the research findings themselves may contain errors or even fraud. The model must not only learn from literature and experimental data but also assess evidence quality, understand disciplinary principles, and establish a validation loop with real-world experiments. It remains uncertain whether each major discipline will develop its own foundational model or whether a unified, cross-disciplinary architecture will emerge. However, the core issue it points to is already evident: as the marginal value of general internet data declines, high-quality data, scientific laws, and experimental feedback from the professional world may become a key driver for the next wave of model capability growth.
Trend Four: The closer you get to the deep waters of hard technology, the more patient capital you need.
Compared to AI applications, the validation cycles for commercial aerospace, quantum computing, space-based computing, and advanced energy are longer. These fields cannot rely on data from just a few months after a product launch to prove their viability; technological pathways, engineering capabilities, supply chains, certifications, industry standards, and market demands must mature together over an extended period.
Wang Shuhe, Vice President of Houxi Capital, exemplifies this restraint in his assessment of quantum computing. Currently, various approaches—such as photonic, neutral atom, and ion-based systems—each have their own advantages and bottlenecks, and there remains significant uncertainty regarding when general-purpose quantum computing will become a reality. Beyond engineering challenges like device design and supply chains, the industry also faces fundamental issues in materials, algorithms, physical mechanisms, talent development, and evaluation standards. A more realistic path may be to enable modular collaboration between quantum computing and classical supercomputers, GPUs, and CPUs, rather than expecting general-purpose quantum computing to independently replace existing systems in the near term.
Uncertainty does not equate to a lack of investment value. Moderate enthusiasm in frontier industries attracts talent, capital, and industrial resources; even if not all companies succeed, the process still yields accumulated engineering expertise, supply chain capabilities, and transferable technological outcomes. The key is that capital must not directly convert long-term possibilities into short-term certainty, nor should it abandon the effort to form early judgments simply because commercialization is distant.
The same applies to commercial aerospace. This industry involves launch operations, approvals, qualifications, testing, quality management, and supply chain coordination—far longer than the typical software product lifecycle. However, its industrialization is accelerating: in 2025, China completed 87 space launches, with private commercial rocket companies conducting 23 of them and placing 324 spacecraft into orbit; total industry funding reached RMB 18.6 billion, a 32% year-over-year increase, and commercial launch service orders grew by approximately 40%. Venture capital cannot wait for companies to fully cross the “valley of death,” as the window for early-stage investment often closes by then. Investors must seek stability amid change: Will market demand ultimately materialize? Does the team possess foundational R&D and engineering capabilities? Can they manage supply chain and commercialization challenges? And do they have sufficient resilience to endure until the industry’s gates truly open?
Wang Shuhe’s focus on embodied space robots is a specific example of this judgment. It sounds like a convergence of several hot concepts—commercial spaceflight, robotics, and AI—but clear demand has already emerged. As the number of satellites increases, managing decommissioned satellites, cleaning up space debris, in-orbit operations, and in-orbit construction will become integral parts of the space ecosystem. When commercial spaceflight shifts from “how to get equipment into space” to “how to sustainably use and maintain space resources,” new infrastructure and service systems will emerge.
The attention given to space computing, aerospace energy materials, and quantum computing on the leaderboard shows that young investors are not limiting their focus to AI applications with quicker returns. They are also entering fields where technological pathways have not yet fully converged but may reshape the foundation of future industries. Patience in hard tech does not mean abandoning returns; rather, it means betting on returns over a longer time horizon. Such investments require capital that is both early and patient.
As AI moves from the cloud to devices, from digital content to physical action, and as tech investments expand from software to chips, energy, robotics, and space infrastructure, the role of capital will also change. It must understand technology earlier, penetrate industries more deeply, and make long-term decisions even when outcomes are uncertain. What makes this list worth watching is how this generation of investors will learn to make such choices.
This article is from Huxiu, authored by Chen Yifan_YF. Original link: https://www.huxiu.com/article/4877273.html?type=text